Elsevier

Energy and Buildings

Volume 258, 1 March 2022, 111829
Energy and Buildings

Simulation-Based Optimization of Residential Energy Flows Using White Box Modeling by Genetic Programming

https://doi.org/10.1016/j.enbuild.2021.111829Get rights and content

Abstract

The development of energy management systems that optimize the electrical energy flows of residential buildings has become important nowadays. The optimization is formulated as a symbolic regression problem that is solved by genetic programming, which provides near optimal results while being highly performant during application. Additionally, the so-trained energy flow controllers are explainable and therefore address three of the current major disadvantages of most existing solutions. 260 controllers are trained to calculate the optimal gridfeed-in value for an inverter and are evaluated for their ability to minimize the energy costs and to support grid stability and battery lifetime. Additionally, they are compared to two existing energy management systems, a rule-based self consumption optimization and a linear model predictive controller. It is shown that this energy management system can significantly minimize energy costs compared to both reference systems by up to 58.25%, support grid stability and prolong battery lifetime by up to 76.48%.

Introduction

In recent years, renewable energy sources have become increasingly important in order to slow down global warming and drive forward energy transition. Therefore, the construction of renewable energy systems like photovoltaic (PV) systems has been strongly promoted and was proposed to reach 38–40% of the final energy consumption share within the European Union by 2030 [1]. However, due to constantly changing environmental conditions, these power plants are not able to produce a constant amount of energy, which results in bigger and faster fluctuations in the low voltage grid. In order to keep the grid stable, these fluctuations need to be handled by the grid operating reserves. At the moment, they can still be compensated well, but in the future with more renewable energy sources installed, this will become increasingly difficult. This is why some countries, e.g. Germany, have already implemented regulations that limit the amount of allowed feed-in energy [2]. Thus it is important to use as much of the self-produced energy as efficiently as possible. As a result, energy management systems (EMS) have become more important and also increasingly prominent in research. Such systems should use, store and distribute the energy produced by renewable energy sources as efficiently as possible in order to minimize the energy costs of buildings. Common EMS in the form of rule-based systems are easy to implement and runtime efficient but not optimal, while the ones based on model predictive controls are computationally intensive and usually not real-time capable but can provide almost optimal results. Model predictive controls usually also lack in scalability because they have to be developed anew once the setup of a system changes.

The main goal of this paper is therefore to present a computationally efficient, highly scalable and near-optimal energy management system for residential households with a PV system and battery storage. For that, the developed self-learning energy flow controllers are trained to calculate the optimal gridfeed-in value for the inverter (Fig. 1) to minimize the system’s energy costs. This is done using symbolic regression performed by genetic programming. Thus, the controllers are very performant and even real-time capable during their application while also being human-understandable and explainable. The controllers can also be easily re-trained in case the system changes, which makes them highly scalable. Besides that, they are highly customizable as their training is done using data that is measured directly from the system to be optimized.

In order to evaluate this approach, 260 energy flow controllers are trained with measured data including variable energy tariffs, household load and PV production as well as the respective forecasts. All these controllers are then evaluated by running them in simulation with different evaluation timespans and comparing them to two existing EMS, which are also evaluated via simulation. The first reference EMS is the Fronius self consumption optimization (SCO) [3], a simple rule-based approach, and the second one is a linear model predictive controller (MPC) implemented by Kirchsteiger et al. [4].

The remaining work is structured as follows: Section 2 gives an overview of existing EMS technologies, followed by a detailed description of the developed methodology in Section 3 and the used data basis in Section 4. Section 5 Evaluation, 6 Results describe the evaluation and its results, while Section 7 contains the conclusions.

Section snippets

Related Work

Currently, four main trends can be identified for the optimization of residential energy flows: Rule-based control systems, model predictive controls, linear programming approaches and meta-heuristic optimization algorithms as used in this work. All approaches are briefly explained below.

Method

The optimization approach is shown in Fig. 1 and is based on a further development of the model-based optimization approach by Kefer et al. [20]. A controller is trained to calculate the optimal grid feed-in value for the inverter as shown in Fig. 3, Fig. 4, so that the energy costs of the system are minimized. For that it uses provided input data about the current state of the system including current energy tariffs, PV production, household load and state of charge (SOC) of the battery. The

Data Basis

As data basis, more than three years of measured data (12/2015–2019) from a single-family household in Upper Austria is used. The climate in this area is cool temperate, i.e. there are four distinct seasons with up to −10°C and snowfall during the winter and up to 30°C during the summer and also varying sunshine rates [23]. The measured data includes the production as well as the voltage of the PV system and the household load. The data was recorded using a Raspberry Pi computing unit that was

Evaluation

For the evaluation of this EMS, five controllers are trained with each genetic algorithm and training dataset, resulting in 260 controllers. The first algorithm is the single-objective Offspring Selection Genetic Algorithm (OSGA) by Affenzeller et al. [25], which minimizes the system’s energy costs. As parameters, a population size of 500, 1000 selected parents, maximum 100 generations and 50 000 evaluated solutions, a mutation probability of 15% and a maximum selection pressure of 100 are

Results

This chapter summarizes the results of the evaluation. Section 6.1 presents the results for the controllers trained without forecasts, followed by the results for the forecast controllers in Section 6.2. Section 6.3 shows the results for the different training data lengths, forecast types and lengths and the computational efficiency, followed by the results of the overall best controller concerning its grid stability and battery lifetime support in Section 6.4. Fig. 6 shows the best energy

Conclusion

In this work a new energy management system that formulates the energy flow optimization as a symbolic regression problem which is solved by genetic programming is presented. With the developed approach, 260 energy flow controllers are trained with measured real world data and two different genetic algorithms to calculate the optimal grid feed-in value for the inverter in order to minimize the energy costs of a residential building with a PV system and a battery. The controllers are evaluated

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

This project was financed by the European Regional Development Fund and the Province of Upper Austria. It was carried out by Fronius International GmbH and partners from the University of Applied Sciences Upper Austria.

References (28)

  • J. Matousek et al.

    Understanding and using linear programming

    (2007)
  • T. Hubert et al.

    Modeling for residential electricity optimization in dynamic pricing environments

    IEEE Transactions on Smart Grid

    (2012)
  • F. De Angelis et al.

    Optimal home energy management under dynamic electrical and thermal constraints

    IEEE Transactions on Industrial Informatics

    (2012)
  • Z. Bradac et al.

    Optimal scheduling of domestic appliances via milp

    Energies

    (2015)
  • Cited by (9)

    • A framework of data modeling and artificial intelligence for environmental-friendly energy system: Application of Kalina cycle improved with fuel cell and thermoelectric module

      2022, Process Safety and Environmental Protection
      Citation Excerpt :

      Luo et al. (2022) also applied multi-objective optimization of the cascade storage system in hydrogen refueling stations intending to minimize cooling energy and maximize state of charge. Kefer et al. (2022) used simulation-based optimization of residential energy flows by making use of genetic programming. Similar research is also available on the usage of data modeling for optimization in poly-generation systems driven by a geothermal energy resource.

    • Assignment of Unexpected Tasks in Embedded System Design Process Using Genetic Programming

      2024, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    View all citing articles on Scopus
    View full text